Open Access

ARTICLE

Deep Learning-Based Glass Detection for Smart Glass Manufacturing Processes

Seungmin Lee¹, Beomseong Kim², Heesung Lee^3,*

1 Department of AI Transportation Convergence, Korea National University of Transportation, Uiwang-si, 16106, Republic of Korea
2 Department of Artificial Intelligence, Gyeonggi University of Science and Technology, Siheung-si, 15073, Republic of Korea
3 Department of Railroad Electrical and Information Engineering, Korea National University of Transportation, Uiwang-si, 16106, Republic of Korea

* Corresponding Author: Heesung Lee. Email:

Computers, Materials & Continua 2025, 84(1), 1397-1415. https://doi.org/10.32604/cmc.2025.066152

Received 31 March 2025; Accepted 16 May 2025; Issue published 09 June 2025

Abstract

This study proposes an advanced vision-based technology for detecting glass products and identifying defects in a smart glass factory production environment. Leveraging artificial intelligence (AI) and computer vision, the research aims to automate glass detection processes and maximize production efficiency. The primary focus is on developing a precise glass detection and quality management system tailored to smart manufacturing environments. The proposed system utilizes the various YOLO (You Only Look Once) models for glass detection, comparing their performance to identify the most effective architecture. Input images are preprocessed using a Gaussian Mixture Model (GMM) to remove background noise present in factory environments. This approach minimizes distractions caused by varying backgrounds and enables accurate glass identification and defect detection. Traditional manual inspection methods often require skilled labor, are time-intensive, and may lack consistency. In contrast, the proposed vision-based system ensures high accuracy and reliability through non-contact inspection. The performance of the system was evaluated using video data collected from an actual glass factory. This assessment verified the accuracy, reliability, and practicality of the system, demonstrating its effectiveness in real-world production scenarios. Beyond automating glass detection and defect identification, the proposed system integrates into manufacturing environments to support data-driven decision-making. This enables real-time monitoring, defect prediction, and improved production efficiency. Moreover, this research is expected to serve as a model for enhancing quality control and productivity across various manufacturing industries, driving innovation in smart manufacturing.

---

Keywords

---